Brain Inspires New Type of Circuits

Researchers at the School of Engineering and Applied Sciences have employed a new material to mimic the low-power, high-performance functioning of the human brain in simple circuits.

By exploiting the properties of correlated oxides—a family of materials that can fluidly change their electronic properties—professor Shriram Ramanathan and coworkers have made a device that can adapt in a way similar to brain cells.

“If an acquaintance comes towards you, you can recognize him or her in less than a second, as the time range for a neuron to work lies in [the range of] milliseconds,” said Jian Shi, a postdoctoral researcher in Ramanathan’s lab.

She said that this type of speed and low energy inspired material scientists.

The brain not only splits computation between many units, but also incrementally strengthens pathways after repeated use.

This is believed to be necessary for learning and memory.

But rather than just turning on or off certain neural connections, the brain increases the ability for a connection to carry an electric signal the more it is used.

This property is mimicked by incorporating correlated oxides into circuit connections.

Ramanathan’s group used samarium nickelate, which incrementally adjusts how well it conducts electricity according to how strong an electric field the circuit machinery applies to it.

As in the brain, these subtle shifts in conductivity are made with low energies.

“One of the key limiters for the functionality of mobile devices is the energy consumption,” Ramanthan said. “You’re basically limited to the power of your battery, so if you want a state-of-the-art computer running on your iPhone, that can be very challenging.”

This variable conductance behavior is at the core of the semiconductor industry, but still now, the industry suffers from the fact that highly structured silicon, which has classically been the best option for semiconducting devices, is quite expensive to produce.

Because the nickelate functions despite high disorder, it could eventually be cheaper to produce than current silicon products.

Plus it would not suffer from memory loss when unplugged from a power source.Ramanathan added that his group is currently attempting to incorporate this component into simple circuits, potentially to elucidate details of how brains affect behavior and perform tasks, such as pattern recognition.

—Staff Writer Manny I. Fox Morone can be reached at mmorone@college.harvard.edu. Follow him on Twitter @mannyfoxmorone.